praivate

 ____      _    ____ _   _
|  _ \    / \  / ___| |_(_)_ __ ___   ___
| |_) |  / _ \| |  _| __| | '_ ` _ \ / _ \
|  _ <  / ___ \ |_| | |_| | | | | | |  __/
|_| \_\/_/   \_\____|\__|_|_| |_| |_|\___|

A complete RAG and knowledge system. A platform for opening up the entire AI system.

praivate is a fully functional, self-hosted retrieval-augmented generation (RAG) and knowledge system in its own right. At the same time, it is a modular platform for advancing toward a fully open AI system: one in which knowledge sources, processing steps, components, evidence, and decisions can be inspected, evaluated, and controlled.

These two purposes reinforce each other. praivate delivers practical knowledge retrieval and evidence-grounded answers, while its explicit capability boundaries create room to replace components, experiment with different approaches, and make the complete path from source material to answer increasingly transparent and reproducible.

Open models and open-source components are important, but they are not enough. praivate treats openness as a property of the whole system, not merely of its individual parts.

What guides the design

• Evidence before assertion. Retrieval relevance alone does not establish that the available material is sufficient to answer a question. The system should distinguish supported answers, partial answers, and abstention.
• Provenance, not just logs. Semantic provenance records source material, transformations, evidence, component versions, configuration, and decisions. Operational tracing serves a different purpose: understanding execution and performance.
• Replaceable capabilities. Ingestion, retrieval, fusion, reranking, and generation are separated by explicit contracts. A particular library, model, or backend should not define the architecture.
• Quality as observations. Data quality is described through distinct, provenance-bearing observations, rather than hidden behind a universal score.
• Bounded work. Resource and stopping decisions must respect an explicit, normalized execution budget (B <= 1). More computation is not automatically a better answer.
• Operator control. Knowledge sources, model artifacts, deployment choices, and external dependencies remain explicit responsibilities of the operator.

These principles describe the intended architecture; they should not be read as a claim that every capability is already implemented or independently validated.

The 1.0 baseline

The first public source release is intended to establish a small, coherent foundation:

• a fresh-install-only baseline with typed configuration and explicit capability boundaries;
• a fixed, end-to-end ingestion-to-answer path over operator-controlled material;
• retrieval, evidence handling, and answer production with recorded provenance;
• separate treatment of data-quality observations, evidence sufficiency, and resource/stopping policy;
• operational observability kept distinct from semantic lineage;
• documented component, model, deployment, and publication responsibilities.

The goal is a usable RAG and knowledge system and a coherent platform for further development—not a universal ingestion framework or an already autonomous AI optimization platform.

Adaptive ingestion, experimental policy optimization, and broader automatic system adaptation are directions for later work, not promises for 1.0.

Models, deployment, and dependencies

praivate is designed for self-hosted operation, including deployments that do not rely on externally hosted model APIs. Infrastructure dependencies and supported offline operating conditions must be documented for each release and deployment profile.

Model weights are not distributed with the application. Operators obtain, configure, and license model artifacts separately. The licensing of project software, third-party dependencies, models, and datasets must be assessed independently.

A public release must also identify its tested platforms, required services, configuration process, and limitations. No installation command or platform-support claim is made here before the corresponding public release has been verified.

Project status

In development — 1.0 is not yet a published, verified source release.

This repository serves as the public-facing home for project documentation and the planned source distribution. Its documentation describes architectural intent and release criteria; it is not evidence that a complete installable release already exists.

The first public source publication is planned as a clean baseline without importing private development history, secrets, internal operational material, model weights, or private datasets.

Documentation

• Architecture: capability boundaries and release requirements.
• Configuration: schema, profiles, and validation requirements.
• Installation: clean installation and reproducibility requirements.
• Validation report: documentation checks and pending release evidence.

License

praivate is licensed under the Apache License, Version 2.0. See LICENSE for details.

Model weights, datasets, and third-party materials may be subject to separate licenses and distribution conditions.

Security and contributions

See SECURITY.md for the security policy and the status of confidential reporting, and CONTRIBUTING.md for contribution guidance. A confidential vulnerability reporting channel is not yet available.